Don't confuse me with common sense, I was just beginning to understand
epidemilogy ;)
-----Original Message-----
From: Andrew Jull [mailto:[log in to unmask]]
Sent: Tuesday, September 21, 1999 12:21 AM
To: [log in to unmask]
Subject: Likelihood ratios
Dear all
many thanks for the responses to my previous question. Several respondents
replied directly to me and I found Alan Hackshaw's reminder that it is the
concept of risk that underpins clinical epidemiology particularly pertinent.
Below is Alan's message. Many thanks.
regards
Andrew
"If the sensitivity and specificity are both 100% then there cannot be
a likelihood ratio, because there is no longer a concept of risk.
People either have a disorder or they do not and the aim is to predict
this. Some tests are not able to determine this with 100% accuracy -
and these will provide a
LR and therefore an estimate of risk ( think of 2 overlapping
Gaussian distribution curves, one for affected individuals and one
for unaffected). If these curves do not overlap at all (so
sensitivity=100% and specificity=100%), then the test
is able to completely determine the disease status and there is no
need for a risk estimate - "risk" for an affected person is always
100% and "risk" for an unaffected person is always 0%) - there are
no risks in between these 2 values because there is no likelihood
ratio."
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|